Solubility Prediction for Carbon Dioxide in Polymers by Artificial Neural Network

Solubility of carbon dioxide in poly(vinyl acetate) (PVAc) at temperatures (313.15373.15K) and pressures up to 17.5 MPa, poly(2,6-dimethyl-1,4-phenylene ether) (PPO) at temperatures (373.15-473.15K) and pressures up to 20 MPa, polypropylene (PP) and high-density polyethylene (HOPE) at temperatures (433.2473.2 K) and pressures up to 17 MPa, poly(butylene succinate) (PBS) and poly(butylene succinate-co-adipate) (PBSA) at temperatures (323.15-453.15 K) and pressures up to 20 MPa and polystyrene (PS) from (373.15-473.15 K) and pressures up to 20 MPa are modelled by artificial neural network (ANN) technique. Different types of neural networks have been employed and it is found that the designed ANN, multi-layers perceptron (MLP) architecture, with one hidden layer can be better trained than other types of neural networks. Suitable activation functions for output layer are linear and for hidden layers are sigmoid. The best ANN model derived, a 2-5-1 architecture, has an average relative deviation (ARO) of 0.721 for the test set of PPO(H), 1.319 for PPO(L), 0.599 for PS, 0.583 for PVAc, 1.535 for PP, 0.635 for PBS, 0.503 for PBSA and 1.490 for HOPE. The results obtained in this work indicate that ANN is an effective method for prediction of solubility of carbon dioxide in polymers and have better speed and simplicity compared with the classical methods.